[RECORDING] Transformation Labs 101: Get executive buy-in on AI priorities

Hi everyone, thank you so much for joining us today for Transformation Labs 101. In the next 30 minutes, we are going to start with a real business priority and help you turn the related work into AI opportunities that you can actually compare, and build a recommendation that an exec sponsor can actually respond to so that you can get buy-in.

I have put a link in chat with an extra bracket at the end—sorry about that—but that is where you can open the worksheet from your confirmation emails if you haven't done so already. It's linked there at the top of our chat, where you can copy or download it from OpenAI Academy. That's what we suggest you use to take notes as we work through today. [LINK: https://academy.openai.com/home/clubs/champions-ecqup/resources/get-executive-buy-in-on-ai-priorities-2026-08-18]

Today begins with one real business priority that you've brought. If you already have possible AI opportunities that you're considering, you can start to list them. If you don't quite have them articulated yet, that's okay. Today will help you identify the recurring work that matters, and two or three candidates is really good to give you the clearest comparison. But you can start with one if that's what you brought today. That's perfectly fine.

You also don't need to finish your polished strategy recommendation today, but you will leave with what you need to make a credible recommendation and a clear sponsor ask. We'll build those things in four steps that we'll call Align, Qualify, Prioritize, and Ask.

I'd like to start with what you're actually bringing into the session today, so I'm going to go ahead and kick off a poll, because I'd like to know where you're all starting as you come into the work today. Choose the answer that is closest to your starting point. It's okay if it's not exact. Your answer is just going to point to the next decision that your work is going to need. You don't have to make that decision right here and now; just capture it.

For example, if you have a business priority but no possible AI opportunity yet, your next decision might be around what recurring work most affects that outcome. If you have an opportunity that's a little bit more specific that you're already considering, it may be whether you can state the work, the friction, and the intended result clearly enough to compare it against other potential AI priorities.

If you have several opportunities you're considering, it might be which should move first and which trade-offs that choice accepts. If you already have a likely first priority, it might just be the exact decision or support that you need from your exec sponsor. Wherever you're starting today, each step will help create the outputs that the next step needs.

The four steps: Align, Qualify, Prioritize, and Ask

Each step here solves a different decision problem and helps produce the inputs for the next step. In Align, we're really answering, “Why is this work worth considering?” This prevents us from starting with an appealing AI idea and inventing a business reason for it afterwards.

You're going to want to start with a real business priority: what matters most to your organization now? Then ask what recurring work affects that outcome. Then you can start to ask what friction in that work today causes delay, rework, inconsistency, or just difficulty. Then you can start to define what observable result would improve if that friction were resolved. We're not choosing an AI solution yet; we're identifying where it may be able to help.

Finally, you'll also want to ask a secondary question around what conditions, like safety or controls or quality standards, can't be compromised as the work changes. All of these pieces become our opportunity statements.

Then, in the Qualify step, we can start to answer, “How do these opportunities compare?” We'll apply the same three lenses to every candidate, or every different AI opportunity that you're seriously considering.

Impact asks how much the opportunity could affect the outcomes that matter, or the business priority. Effort asks what the opportunity would take to test and introduce responsibly. That's going to include readiness, things like a defined process, clear ownership, and safeguards—if those exist or do not yet today—and the necessary technical and enablement and support capacity to actually implement the solution in your organization for its intended users.

Finally, confidence asks how strongly the impact and effort judgments that we've made are supported by known facts or direct observations, rather than assumptions or estimates that lack strong evidence at the moment. The output is really just one comparable profile per candidate or AI initiative that you're considering.

We'll cover each element in more detail when you build out those profiles a little later today. But this step really helps prevent us from comparing a really well-developed idea and an undeveloped one that simply sounds good, and pretending it's a meaningful comparison.

Then Prioritize helps us answer what we should explore first and what trade-off we're accepting, or at least making that visible. You're going to choose First using the AI priority with the strongest impact-to-effort trade-off and enough confidence to justify a bounded first move.

You can bound this work by the users or work types included—for example, routine work as opposed to more complex exceptions—the time period that you're going to test it within, and what specific decision it'll help inform about moving that AI initiative forward or expanding it further.

You'll want to put an opportunity in Next when the impact or effort trade-off is sound, but a material gap—a significant, specific condition—needs to be resolved before it can move forward. You'll want to name exactly what must become true, such as a named review team having capacity in the next planning cycle, or an existing stable process being standardized across teams, even if that change can't realistically happen today.

You'll want to put an opportunity in the Later bucket when moving that opportunity forward would require first establishing, or fundamentally changing, a basic prerequisite. For example, maybe the organization has not yet decided what the underlying process should be, or the required data isn't collected or can't be used under the current policies today, or no function really has authority to own the workflow yet. That hasn't been defined.

You'll want to state that foundational condition that's missing, and that way you have a record of what would trigger revisiting or reconsidering that opportunity. The output to this First, Next, Later categorization is really an adaptive sequence. We want to emphasize adaptive here because this prevents your roadmap from becoming a static ranking that never changes, even when the facts change.

Finally, Ask is where we take all of this and clearly state what decision will move the first priority forward. This is where you'll state your recommendation, including which AI opportunity you recommend pursuing first, the business result it connects to, and why we should pursue it now.

You'll want to include the evidence that helped inform your confidence, and what open questions or unknowns or risks might affect that confidence the most, for visibility. And, of course, the specific decision or support or resources needed that you need the exec sponsor to approve.

That way you're giving your exec sponsor the context they need and a specific decision to approve, revise, or reject. This just prevents the strong analysis you've done from ending without giving the decision-maker something really concrete to decide.

Find your earliest missing output

Of these four inputs, take a look and start to identify the earliest output towards the bottom of the screen that your work is still missing at the moment. That could be the opportunity statements, those comparable profiles, that adaptive roadmap sequence, or a decision-ready recommendation.

I promise we don't have too many polls before we dive into the example and your worksheet a little bit further today. But I'm going to launch one more, just to get a sense of what first output your recommendation still needs. You'll want to choose the earliest missing step or incomplete step.

Align is the one that should leave you with clear opportunity statements tied to a business priority—a real one that matters at your organization now. Those statements become the candidates that the Qualify step compares.

Qualify should leave you with a provisional impact, effort, and confidence judgment supported by evidence and explicit assumptions. That comparison is what becomes the input you need for the Prioritize step. We'll dive into each of these steps in more detail. Right now we're just noting where you're at in the work again, so that you know what to focus on as we walk through your worksheet today.

Prioritize should leave you with that First, Next, and Later, as well as the trade-off behind the sequence you've recommended, and the revisit trigger for the Next and Later opportunities. That sequence really becomes the core of your roadmap recommendation.

Ask is what turns that sequence into a clear recommendation with a specific decision, support, or resource requested. That becomes your ask for your exec sponsor.

Just note what's missing at the earliest step that you haven't quite completed yet, and, if it's relevant at this point, who might be able to provide that evidence or decision if it's someone other than you.

Northfield Mutual: Compare opportunities

What we'll do now is apply this method to Northfield Mutual. This is a fictional insurer reviewing a use case where AI might support claims operations. The business priority right now for Northfield is improving policyholder retention.

We're going to work through the routine correspondence example in more detail so that you can see the method applied in practice. But the other examples still compare the candidates on the same criteria: impact, effort, and confidence.

In this example, the recurring work is drafting routine claims correspondence, and the main friction is the time handlers spend drafting and revising really similar messaging. The result that would improve if this friction is resolved is, of course, reduced drafting time. The condition that really needs to be preserved, or that we can't compromise on at Northfield, is quality.

Remember, the business priority is improving policyholder retention. Here our opportunity statement becomes: reduce time drafting claims correspondence so policyholders receive faster, clearer communication while preserving quality. That's all in service of improving policyholder retention.

The impact basis here is strong. The work is high volume, so our frequency is high. It follows a repeat pattern, so it's quite repeatable, and connects directly to policyholder service and retention. The effort basis is moderate. Human review already exists in this process, with very clear ownership around the claims managers. But information spans across several different systems that we would need to access, and governance and capacity, of course, are still needed to redesign this process.

Our confidence is medium. The volume and the delay are really easy to directly observe, but the belief that AI-supported drafts will actually reduce work remains an assumption that we need to test.

For the other two examples, the first being internal meeting summaries, the opportunity is to reduce missed follow-ups so cases move with fewer delays while preserving decision-makers and owners. Impact here is weak because the connection to retention is still unclear. Effort is low because the workflow is simple, mostly internal, and reversible.

Our confidence here is mixed. The workflow is visible, but the idea that saved time is actually going to shift to higher-value claims work is an assumption that we're making.

For complex claims recommendations, the opportunity is to reduce inconsistency so that decisions are better supported while accountability is preserved. Impact is moderate. These decisions are consequential, but the work is lower volume than routine claims correspondence, for example.

Effort here is higher, though, because this work varies a lot. It uses sensitive information, has a lot of dependencies, and would require some pretty extensive testing and approval for us to be able to trust it. Our confidence here is weaker because the expected decision-quality benefit that we would be looking for is still mostly an assumption.

Northfield now has three comparable profiles. They have not tested every assumption or chosen a winner yet, but they have enough consistent information to move to the Prioritize step, which it sounds like a small majority of us are working through today, so we'll spend some time here.

Northfield Mutual: Choose First, Next, and Later

In our example, Northfield decided to recommend routine correspondence First. It has the strongest combination of impact, manageable effort, and enough confidence to justify moving forward.

For Northfield, a bounded first move means routine correspondence only, selecting claims handlers as a small pilot group, a six-week test, and a decision at the end to continue, revise the AI workflow that we design, or pause. That doesn't guarantee success, but it limits the first commitment while Northfield answers this key open question that we have: will a dependable first draft actually reduce drafting time and rework without weakening quality?

Routine correspondence is the strongest responsible place to begin while staying connected to our business priority of increasing policyholder retention.

They sequenced internal meeting summaries Next. The workflow was really straightforward and reversible, so that effort is low. The reason it doesn't move first is impact. Northfield doesn't really know that meeting-summary time savings is going to materially improve the claims experience or retention.

We want to reconsider this one if evidence shows that fewer missed follow-ups meaningfully improve claims progress the way we are right now assuming they will, or if the business priority changes and this is a closer connection to that.

Complex claims recommendations belong in Later for Northfield. The potential value may be meaningful, but the work is pretty variable and high consequence, and our confidence is lower here. Additionally, the effort to establish information access, controls, testing, and approvals is pretty high. We're deliberately deferring this for now.

Sometimes at OpenAI we call these strategic initiatives. They could have a lot of upside and value, but there's also going to be a lot more complexity or effort to implementing those. These can be really great when you've established momentum. But right now, for Northfield, our routine claims correspondence is going to be a more streamlined implementation because of the higher confidence and lower complexity. That'll help create the momentum we may need around a more strategic or larger and more complex initiative later.

The accepted trade-off here is pretty clear. Northfield's not choosing the easiest idea, and it's not choosing the idea with the biggest theoretical upside. It's choosing the strongest responsible starting point.

Finally, the revisit trigger just keeps the roadmap adaptive. Northfield will review and may update the roadmap when evidence, capacity, dependencies, business conditions, or risk changes. Your sequence isn't a commitment to deliver all three of these solutions in this order. It's a means of managing your portfolio of AI opportunities by prioritizing those with meaningful value and realistic implementation effort according to the current conditions of the business.

Build your opportunity statements and profiles

Northfield showed us three completed profiles. Now it's time to align and qualify the ideas you brought today. Please feel free to start drafting sections one and two in your worksheet now.

Starting with Align, remember to start with the business priority first: the consequential outcomes that leaders are trying to improve. Then ask which recurring work materially affects that outcome. For each candidate, you'll want to find the main friction. Where does that work wait, repeat, break down, or vary in a way that matters?

If you're close enough to the work that you know that, you can write down the main friction now. If you need to make an educated guess, that's okay. But either way, always validate your assumptions with the people who actually perform the work. Otherwise, you could be inventing a problem to solve.

Next, we'll want to name the meaningful result that should improve if that friction changes. You'll want to note that, as well as any conditions that must remain true. Conditions might be things like required human approval, privacy constraints, accuracy, or another boundary that the organization can't trade away, even if the way the work gets done changes.

Once those pieces are clear, you can start to assemble the opportunity statement. This should give you two or three places where the work could improve to support a business outcome that matters without having to choose an AI solution just yet. We haven't actually gone into needing to design the solution yet. We're just identifying the meaningful places where redesigning the work could help.

Then you can qualify every candidate on the same basis once you have those opportunity statements. Impact asks, “How much could this impact what matters most now?” Look at how often the work happens, how consistent the process is, how many people it reaches, and how directly the result supports the business priority.

Effort asks, “What would it take to test and introduce this responsibly?” Consider things like process complexity and systems dependencies. Readiness is one important part of effort: comparing what the AI opportunity would probably require with what actually exists today.

By that, I mean: is the process clear and repeatable already? Can you name a responsible owner for this workflow or process? Is there capacity to do and support the work? Can the necessary controls and approvals be put in place with current policies?

A workflow can look technically simple and still require substantial effort if ownership, review, or capacity are missing. That's why it's useful at this stage to just size these conditions. You don't need to validate every single input and detail, design the full test, or solve every gap before you compare these candidates. But you want a sense of where each of your opportunities lies on this impact and effort scale.

Confidence asks, “How strong is the basis for my impact and effort judgments?” Here we just want to make sure we're separating known facts and direct observations from assumptions that you would still need to check with the people who do the work, for example, or the people who monitor it.

Evidence supports a judgment. An assumption explains why confidence is limited and how the judgment might change if that assumption proves false. The output is really just one comparable profile for each of the different candidates that you're considering. We don't have a winner yet. We'll use those profiles to prioritize next.

Before you rely on them for comparison, just do a quick check that your opportunity statements are specific enough. Nando's ahead of me in the chat here, but choose one statement and make sure that you can point out the recurring area of work, the current friction, the meaningful result that should improve—and ideally it's observable and measurable—and one condition the organization can't trade away.

The statement should make the scope of the work and the value really clear without trying to design the whole solution yet. You'll want to keep the condition brief. It's one important boundary; you don't have to try to draft the complete set of controls around the work yet.

Make sure you've written at least one opportunity statement in your worksheet. Sharing it in chat is optional. We would love to see what you're working on if you do choose to share. Just remember to amend any confidential or sensitive details that you're not able to share here.

For Northfield Mutual, for example, in claims correspondence, we want to reduce manual drafting so that policyholders receive faster, clearer communication while preserving handler approval.

If your statement seems too broad, you can narrow the area of work. “Improve customer service,” for example, is pretty broad. “Reduce manual drafting of routine account responses” is specific enough to assess for impact, effort, and confidence.

Prioritize and make the sponsor ask

Northfield showed us that completed sequence as well. Let's slow down those final two steps while you carry your own candidates forward. Now we'll use sections three and four of the worksheet.

Prioritize starts with a comparison. Choose First by looking at the strongest impact and effort trade-off with enough confidence to move forward. Then you can state the trade-off really plainly: what makes this candidate worth starting with, and what limitation or uncertainty are you accepting if you move forward with it?

Then you can scope your next move. You can name the users or the types of work included—for example, senior claims managers only or routine cases only—the time period—for example, Northfield's doing a six-week pilot—and the decision the work should inform. Namely, that's going to be whether to expand, revise, or pause investing in this AI opportunity based on what you see.

You're defining the first commitment here, not necessarily designing the entire implementation plan. Then place another valuable opportunity in Next when you can name what needs to be resolved before it can move forward. This is probably a significant but addressable readiness gap, like a technical partner's capacity, for example.

You'll want to put an opportunity in the Later category when you're deliberately not prioritizing it or deferring it right now, and record why. A revisit trigger is a specific change that should make the team compare the order again—for example, maybe new evidence or a change to business priorities or a new risk signal that surfaces. That's what keeps the sequence adaptive.

Once you've got your sequence settled, you can build your ask. You'll start with your recommendation: which opportunity did you place First, and what business result does it support? Then give the basis for that recommendation: your impact and effort evidence and your level of confidence, as well as any key assumptions or open question.

You'll state the condition from Align that needs to remain true, the boundaries of the first move, and why the other opportunities are Next or Later. Finally, you can name the executive decision that you need. Ask the sponsor to approve, revise, or reject the sequence that you're recommending and the first move, and also clearly state the support that's needed, whether it's a named owner's time, reviewer capacity, governance support, technical help, or some combination of those.

For Northfield, that becomes a six-week routine correspondence pilot with claims handler approval preserved, plus a request for the senior claims manager's time needed to begin.

Use ChatGPT to prepare the brief

Of course, once your worksheet reflects the judgment, ChatGPT can help you turn it into a concise leadership brief. Make sure you use only information you're permitted to enter into a tool that's approved by your organization. We do have a prompt at the end of your worksheet to help you with this.

You can provide your worksheet itself and your notes as context and ask ChatGPT to organize each opportunity into the same impact, effort, and confidence fields, separate evidence from assumptions, and flag any unsupported claims or things that you still need to validate as “needs input.” It can help you draft the complete First, Next, Later sequence and your decision-ready recommendation. It can also show where two opportunities were described at different levels of detail if you need to correct that comparison, for example.

AI can structure and summarize. You're still accountable for the judgment call. You decide the business priority, the impact and effort trade-off, your confidence level, the sequence, the conditions that need to hold, and the specific ask. Always make sure to check every fact before you share the draft.

If you can use your approved tool now, run the prompt from the worksheet while we wrap up. But if not, you can take that as your next action, as you'll be walking away with the worksheet today.

Next steps

Before you leave today, make sure you've started to capture your sponsor by naming them and your ask for them. I always recommend setting a specific timeline to make that ask or scheduling when you're going to make that ask. Then you can update your roadmap sequence based on their decision.

Additionally, if you are a customer on our OpenAI Enterprise plan, you can apply to the OpenAI Enterprise Champion Network at champions.openai.com. Once you've been accepted, go ahead and sign up for our monthly Champion Roundtable to compare your work and work through any questions you have or challenges as you're working through your recommendation with other Champions.

You can also apply for our Champion Network badges, like the AI Roadmap Foundations badge in the Champion Network, when you register for roundtables or just share your work with us there. To earn that badge, you'll just share your real example. Make sure to omit any details you're not permitted to share. But it won't be shared any more broadly than the network without checking with you first.

We'll look for things like that first opportunity, its connection to a real business priority, how you're going to measure the meaningful result, your impact and effort trade-off, your confidence, and your exec decision-ready ask.

Schedule the sponsor conversation, apply to the Champion Network, and register for the next roundtable if you're eligible. Thank you so much for building alongside me today. I can't wait to see what you lead.